A NNALI DELLA
S CUOLA N ORMALE S UPERIORE DI P ISA Classe di Scienze
R. D IECKERHOFF
E. Z EHNDER
Boundedness of solutions via the twist-theorem
Annali della Scuola Normale Superiore di Pisa, Classe di Scienze 4
esérie, tome 14, n
o1 (1987), p. 79-95
<http://www.numdam.org/item?id=ASNSP_1987_4_14_1_79_0>
© Scuola Normale Superiore, Pisa, 1987, tous droits réservés.
L’accès aux archives de la revue « Annali della Scuola Normale Superiore di Pisa, Classe di Scienze » (http://www.sns.it/it/edizioni/riviste/annaliscienze/) implique l’accord avec les conditions générales d’utilisation (http://www.numdam.org/conditions). Toute utilisa- tion commerciale ou impression systématique est constitutive d’une infraction pénale.
Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
R. DIECKERHOFF - E. ZEHNDER
1. Introduction and results
a) -
ResultsIt is well known that the
longtime
behaviour of atimedependent
nonlineardifferential
equation
f being periodic in t,
can be very intricate. Forexample,
there areequations having
unbounded solutions but withinfinitely
many zeroes and withnearby
unbounded solutions
having randomly prescribed
numbers of zeroes and alsoperiodic solutions,
see V.M. Alekseev[8].
K. Sitnikov[7]
and J. Moser[4].
Incontrast to such unboundedness
phenomena
one may look for conditions on thenonlinearity,
in addition to thecondition,
thatwhich allow to conclude that all the solutions of the
equation
are bounded. Forexample
every solution of theequation
p(t + 1) = p(t) being continuous,
is bounded. Thisresult, prompted by questions
of J.E. Littlewood in
[2],
is due to G.R. Morris[1].
Our aim is to extend theresult to a more
general
but still very restricted class ofequations
for which inparticular
thetimedependence
is involved in thenonlinearity
in x.Pervenuto alla redazione il 9 Gennaio 1986 ed in forma definitiva il 31 Gennaio 1986.
(*)
Supported by
theStiftung Volkswagenwerk.
THEOREM 1.
Every
solutionx(t) of
theequation
with +
1)
= and pj ECoo,
isbounded,
i.e. it existsfor
all t andWe shall not make the bound
explicit
in terms of the initial conditionsx(O)
and£(0)
of the solution. Theproof
of Theorem 1 isstrictly
2-dimensional and based on J. Moser’s twist theoremsimilarly
to theproof
of Morris’ result. Theunderlying
idea is as follows.By
means of transformationtheory
theequation
is outside of a
large
disc D in the transformed into a Hamiltonianequation having
thefollowing property. Following
the solutions from the section t = 0 to the section t = 1 defines a map, the time 1map ~
of theflow,
whichis close to a so called twist map D.
By
means of the twist-theoremone finds
large
invariant curvesdiffeomorphic
to circles andsurrounding
theorigin
in the(x, x)-plane. Every
such curve is the base of a timeperiodic
andunder the flow invariant
cylinder
in thephase
space(~, x, t) x 3,
which confines the solutions in its interior and which therefore leads to a bound of these solutions. It turns out that the solutionsstarting
at t = 0 on the invariantcurves are
quasiperiodic.
THEOREM 2. There is a
(large)
w* > 0 suchthat for
every irrational numberw > w*
satisfying
for
allintegers
p and with two constants,Q
> 0 and c >0,
there is aquasiperiodic
solutionof (1.4) having frequencies (w, 1 );
i.e. there is a smoothfunction F(OI, 02) periodic of period 1,
such thatare solutions
of
theequation.
It is well known that the invariant curves
guaranteed by
the twist-theorem lead to an abundance ofperiodic
solutions. Forexample by applying
thePoincare-Birkoff fixed
point
theorem to the annuli boundedby
two suitableinvariant curves one finds fixed
points
andperiodic points
of the time 1 map of the flow.They give
rise to forced oscillations and subharmonic solutions of theequation.
Oneknows,
inaddition,
that these invariant curves are in the closure of the set ofperiodic points.
Since in our case the set of invariant curves has infiniteLebesgue
measurein J? 2 B D
we shall conclude:THEOREM 3. For every
integer
m > 1 there areinfinitely
manyperiodic
solution
of (1.4) having
minimalperiod
m. Moreover the closureof
the setof
subharmonic solutionsof (1.4)
isof infinite Lebesgue
measure in thephase
x
S’ 1.
For related results
concerning infinitely
many forced oscillations wepoint
out
[5]
and[6].
We shouldmention,
that up to now there is nogenuine generalization
of the firstpart
of this statement tohigher dimensions,
i.e.equations
x =JV h(t, x),
xE ’~2n,
n > 2. In thespecial
case ofequations
ofthe form i =
JVh(x)
+p(t) Berestycki
and Bahri[13]
foundinfinitely
many forced oscillationsusing
minimaxtechniques,
under suitableassumptions
on theHamilton function h.
b) -
Remark about the smoothnessrequirements.
It is
likely
that in thegeneral
case(1.1)
the boundedness of the solution is related to an excessive smoothnessrequirement
in the x-variable. The role of the smoothness in the t-variable is not clear. In thespecial
case of(1.4)
wherethe time is not involved in the nonlinear term, the
dependence
on t isonly required
to be continuous. In fact we shall prove:THEOREM 4.
Every
solutionof
with a continuous
p(t + 1)
=p(t)
is bounded. Moreover the statementsof
Theorem2 and Theorem 3 hold true.
In this statement,
conjectured already
in[ 1 ],
thetimedependent
term isbounded and is
required merely
to be continuous. In contrast, theproof
of thegeneral
case in Theorem 1 with thetimedependent
termunbounded, requires
an excessive amount of derivatives in the t-variable. In fact the number of derivatives in t we need
depends
on the size of thet-dependent
nonlinear term.It will follow from the
proof
that the statement of Theorem 1 holds true forthe
equation
°0 .~
2n,
under thefollowing
smoothnessassumption
which is weaker thanin Theorem 1: pj E where v =
vel, n)
is the smallestinteger satisfying
It is not clear whether the boundedness
phenomenon
is related to the smoothness in the t-variable or whether thisrequirement
is ashortcoming
of ourproof.
We
point
out that relatedproblems
for theequation x
+ = 0 have beeninvestigated by
Coffmann and Ullrich[ 17].
We should remark that the statement of Theorem 1 holds true for a moregeneral
class ofequations
as can be seenfrom its
proof.
2. - Action and
angle-variables
Dropping
thetimedependent
termequation (1.4)
becomes x +x2n+l =
0.Introducing
x = y we have vectorfield x = yand -x2n+’
which is a time-dependent
Hamiltoniansystem
onClearly h
> 0 onR2 except
at theonly equilibrium point (x, y) = (o, 0)
whereh = 0. All the solutions of
(2.1 )
areperiodic,
theperiods tending
to zero as h = Etends to
infinity.
Infact,
since the Hamiltonian function ishomogeneous
in the x-variable,
the solutions areeasily
described in terms of asingle
reference solutionfor which we take
(x*(t), y*(t)) having
the initial conditions(~* (o), y*(O) = (1, 0).
It has the energy , Let T* > 0 be its minimal
period
and introduce the functions C and ~S’
by
’
These
analytic
functionssatisfy
Let r > 0 and set 1
:- 1 ,
then the solution of~ 2.1 )
with the initial conditionsn
(x(O), y(0)) = (r~’, 0)
isgiven by (x(t), y(t)) =
as onereadily
verifies. It has
period
T In terms of its energy wefind for T =
f (E)
the functionf (E)
= C.E-11 with a =n . Loosly speaking
2(n + 1 )
the solutions
spin
around faster and faster withincreasing
energy, whichhas,
as it turns out, a
stabilizing
effect.The action and
angle
variables are now definedby
the )R 2 B {0},
where(x, y) = 0)
with A > 0 and with 0(mod 1)
isgiven by
theformulae:
We claim
that 0
is asymplectic diffeomorphism
from R+ xS
1 onto~~ 2 B 101-
Indeed,
for the Jacobian Aof 0
one findsby (ii)
and(iii) 1,
so that1/J
is measurepreserving.
Moreover since(S, C)
is a solution of a differentialequation having
T* as minimalperiod
one concludesthat 0
is one to one andonto, which proves the claim.
In the new coordinates the Hamiltonian function
(2.1)
becomeswhich is
independent
of theangle
variable 0 so that thesystem (2.1 )
becomes verysimple:
The full
equation (1.4)
has the Hamiltonian function:Under the
symplectic transformation y
it is transformed into:where G is of the form:
with
Gj
ECOO(T2)
andT2
=L~ 2/22.
The Hamiltoniansystem Xh,,
is now morecomplicated:
In the
following
we shall transform thissystem,
forlarge A,
into asimpler
system
for which the0-depending
terms aresmall,
so that in the new variables A is close to anintegral.
To this effect an iterative sequence offinitely
many canonical transformations of R+ xS’
1 whichdepend periodically
on time t will be carried out in the next section.3. - More canonical transformations
First we introduce a space of functions
fer)
which behave forlarge A
>0,
with all theirderivatives,
likeArf,(O,t).
Given r E R we denoteby fer)
theset of C°° functions in
(A, 0, t)
xT~
which are defined in A >Ao
for someAo
> 0 and for which there is a sequenceÀjlk
> 0 such thatWe summarize some
properties readily
verified from the definition:LEMMA 1.
For
f
ET(r)
we denote the meanvalue over the 0-variableby [ f ] :
If
Ao
>0,
thenAao
C R+ xT2
denotes the annulusPROPOSITION 1. Let
with
h,
Efee)
andh2
Ef(b).
Assume a >1,
b a and c a. Then there is a canonicaldiffeomorphism 0 depending periodically
on tof
theform
with u c
~(1 - (a - b))
and v E.7(-(a - b))
such thatA,,,
C CA~,_ for
some
large
J-Lo Moreover thetransformed
Hamiltonianvectorfield
~b* (XH)
=XH
isof
theform:
where
hi
EF(cl),
with cl =max{c, bl,
isgiven by
and where
h2
E1(bl).
The constantbl
is smaller that b and isgiven by
PROOF. The
proof
will follow from several Lemmata. We shall look for therequires reansfonnation 1b given by
means of agenerating
function0, t),
so that
1/Jis implicitely
definedby
In the
following
we shall often supress the variable t in the formulae.Abbreviating:
we have for the transformed Hamiltonian vectorfield =
fl expressed
in the variables
(ti, 0)
instead of(IL, 0):
By Taylor’s
formula we can writewith
where ’ stands for the derivative in A. We now determine v from the
equation
so that
and therefore
Consequently:
LEMMA 2. Let S be
defined by (3.7)
and(3.6).
Then theformula (3.1) defines
a
symplectic diffeomorphism 0 depending periodically
on time tof
theform
PROOF. We first claim that
Indeed,
since a > c wefind,
for 1Lsufficiently large,
and therefore
by
LemmaThen in view of Lemma 1
(iii)
91 - g2 = v E~( 1 - (a - b))
and the claim(3.9)
follows. Next we solve thesecond
equation
of(3.1)
for 0 and writeFor the inverse we set
In view of
(3.10)
we find for v theequation
If p
islarge, then
so that v isuniquely
determinedby
the contractionprinciple. Moreover, by
theimplicit
function theorem v E for somelarge
po. We claimIndeed, apply
to theequation (3.11).
Theright
hand side is a sum ofterms
s
with 1 s + k k. The
highest
order term is the one~ k=1
with s = 0 and k =
0, namely
which weput
to the left hand side of theequation. Inductively assuming
thatfor j
n - 1 the estimatesI
hold true we conclude the same estimatefor j =
n, since g E3~(-(a - b))
and therefore The claim(3.12)
follows.We next insert 0
= Q + v
into the firstequation
of(3.1 )
and define u to beSince v E
7’(1 - (a - b))
in view of(3.9)
and since v E~(- (a - b))
in view of(3.12)
one concludesusing (3.13)
that u E~( 1 - (a - b)).
To finish the
proof
of the Lemma one verifieseasily
that themap y
has aright
inverse of the sametype as 0
defined onA~_
for somelarge
1L- and that it isinjective
on for polarge
For the transformed Hamiltonian function
ÎI,
nowexpressed
in the variablesJ-L, cP
we have in view of(3.8), (3.5)
and(3.14):
where R =
R,
+R2
+R3,
withLEMMA 3.
PROOF.
ad(i)
,S E~(1 - (a - b)) by (3.9)
and v E.1( -(a - b)) by
Lemma2,
henceRl
E~( 1 - (a - b)).
ad(ii)
Setf
= thenf
E7(a - 2). Setting w
:= ~ + Tu, D =D,,
thenis a sum of terms:
s
with 1 n amd 1 s n. Since
f
EF(a - 2)
andk=l
u E
~( 1 - (a - b))
wehave Moreover, if j =
1 then2
for 1L sufficiently large,
since Du E~(2013(a - b)). If j
>1,
thenDjw
=TDJU
E~(1 2013 (a - b) - j).
Therefore the above term is estimatedby J-La-2-n.
Hence gi:= f (w)
EF(a - 2).
Since g2 .u2
E~(2 - 2(a - b))
weconclude in view of Lemma 1
(iii)
that 91 - 92 CF(b - (a - b))
as claimed.ad(iii)
Recall thatDn( f (A, B))
is a sum of termswith and and Since
and one concludes that
Therefore as claimed.
In view of Lemma 2 and Lemma 3 and in view of
(3.15)
theproof
ofProposition
1 is finishedsetting h2
= R.4. - Proof of the Theorems 1-3
If H satisfies the
assumptions
ofProposition
1 then for anygiven
numberd 0 there is an
integer j = j(a, b, d)
so thatafter j
successiveapplications
ofthe
proposition
we find that thecorresponding perturbation
termh2 belongs
towith bj
d. We shall make thenumber j
ofsteps
neededprecise
for ourspecial
case(1.7),
which inangle
and action variables is a Hamiltoniansystem
of the form:where - - and
a = 1
* Therefore 1 a 2 and b a so that the n+2assumptions
ofProposition
1 are met.PROPOSITION 2. Let H be as in
(4.1 ).
Then there is a canonicaltransformation, periodic
in time t :V) =
withA,,,
C CA,,- for
some IL- u+, whichtransforms
the Hamiltoniansystem
into~*(XH)
=HH,
wherewith
hl
EY(b)
andh2
E~(-E) for -
> 2 - a. Thenumber j of transformations
is smaller or
equal to j* = j * (n, t) where j*
is the smallestinteger
>n (.~
+3)
in the case that f n +
1, respectively
in the casePROOF. +
1,
so that b =a(.~ + 1)
1.Then j applications
ofProposition
1 lead to aperturbation
termh2 E 7’(bj) with bj =b-j(a- 1) =
a(£ + 1) - jan,
which is smaller than -2a> ~(£ + 3).
On the othern
hand,
> 1. Set £ + 1 = 2n + 2 - s, then after rapplications
ofProposition
1 wehave br
=a(2n
+2) -
2r sa aslong
asbr >
1.Now br
> 1 if andonly
if 2rS n. Let k be so that2ks
n2k+1s.
Thenbk+l
= 1 and thereforebk+l+q
=bk+l -qan,
hencebk+l+q
-2ahence in
particular
if We find in this casehence the result,..
Let now H be as in
(4.2)
i.e.on for some
large Ao,
withh,
ET(b)
andh2
ET( -6)
for 6 > 2 - a. TheHamiltonian equations
areSince
h2
E~(-~)
one verifieseasily
that the solutions do exist for 0 t1,
if the initial valueA(O) = A
issufficiently large.
In fact we shallconclude
that the time 1 map is close to a twist map.’
LEMMA 4. The time 1 map
ø1 of
theflow Ot of
thevectorfield XH given by (4.3)
isof
theform:
with Moreover
for
everypair (r, s):
PROOF. Set
and set for the flow
(A(t), 0(t)) = 0)
with00
=idThen the
integral-equation
for the flow is
equivalent
to thefollowing equations
for A and B:One verifies
easily
that for A >Ao
theseequations
have aunique
solution inthe
space IAI,
1using
the contractionprinciple,
moreover A and B are smooth. Therequired
estimates can theninductively
be verified from(4.6)
inview of
l(b),
henceaÀ2hl E a2 1(b-2),
andh2 E ~’(- ~). ·
For the
completness
we include aproof
of thefollowing
well known fact(see
f.e.[14]).
LEMMA 5. The
map 7 = ~~
has the intersection property onAÀo’
i.e.if
C is an embedded circle in
AÀo homotopic
to a circle A =const inAÀo’
thenPROOF.
Since 7 = 01
1 is the time 1 map of an exact Hamiltonianvectorfield, XH
it is exactsymplectic,
i.e.l*w-w
= dW for some W onAAO,
where w = Ad0.In
fact,
sinceØO
=id:where ,
Therefore,
if ; 1 is theinjection
map, we have and hence
Assume now,
by contradiction,
that7(C)
n C =0,
then7(C)
U C bounds anannulus in which has
positive
measure.Consequently, by
Stokesformula,
in contradiction to
(4.7).
This proves the LemmaFinally,
in the new coordinatesthe time 1
map ?
isexpressed
as follows:From Lemma 4 one concludes the
following
estimates for theperturbation:
if IL >
s). Therefore,
if it issufficiently large,
themap f is,
with itsderivatives,
close to a standard twist map.Moreover,
it has the intersectionproperty
in view of Lemma5,
so that theassumptions
of the twist-theorem[3]
are met.
It follows that for w >
w*,
W*sufficiently large,
andfor two constants
,C3
> 0 and c > 0 and for allintegers
p andq fl
0 there isan
embedding 0 : S 1 -+
of acircle,
which isdifferentiably
close to theinjection map j
of the circle(w)
X,S 1 --~
and which is invariant under the map,~. Moreover,
on this invariant curve themap ~
isconjugated
to a rotationwith rotation number w :
The solutions of the Hamitonian
equation starting
at time t = 0 on this invariantcurve determine a
1-periodic cylinder
in the space(p, 0, t)
E x ~ . Sincethe Hamiltonian vectorfield
XH
istimeperiodic,
thephase
space is xS’ .
Let
(DI
with =id be the flow of thetime-independent
vectorfield(XH,1 )
onAlto
xS’
1 and define the embedded torus T :T 2 ~
xS’
1by setting
In view of
(4.12)
we havewith # = $ ’
indeed‘~’(s + 1, T) _ ~’( 1, T + 1 ) _ ~’(s, T).
Moreover o
T)
+wt,
T +t),
so that the torus~’(T2)
isquasiperiodic
having
thefrequencies (w, 1).
This proves the statement of Theorem 2. In order to prove the statement of Theorem 1just
observedthat,
in theoriginal coordinates,
every
point (x, y)
ER 2
is in the interior of some invariant curve of the time 1 map of the flow which goes around theorigin.
Its solution is therefore confined in the interior of the timeperiodic cylinder
above the invariant curve and henceis bounded. This ends the
proof
of Theorem 1. ·We next prove Theorem 3.
Following
at first thearguments
of G. Morris[1]
we shall establish fixedpoints
of the iterated map-~m
for everyinteger
m
applying
the Poincare-Birkoff fixedpoint
theorem[9]
to the map#’~
in the annulus R bounded
by
two invariant curves with rotation numberssatisfying (4.11).
First we map this annulus R onto an annulusRo = {(ç,1])IO ~ 1, ~
mod1}
boundedby
concentric circles.If
wj
are theembeddings
of the invariant curves(see (4.12)) having
rotationnumbers wj
we define the map X : Rby
Since the
embeddings 1/;j
aredifferentiably
close to theinjections
of the circleswj =const, x is a
diffeomorphism.
The induced map G :X- I o Fo
x,expressed by
satisfies in view of
(4.12) f (~ + 1, r~)
=f (~, n)
andg(~ +
=g(~, 1]).
It leavesthe boundaries q = 0 and q = 1 invariant and satisfies on these boundaries:
Define the map s :
Ro - Ro by
=(~
+1,7y).
To find aperiodic point
ofminimal
period q >
1 ofG,
for anygiven integer
q, we let[qw]
be theintegral
part
of qw, i.e. qw - 1[qw]
qw, and choose aninteger
0p
q such thatp + [qw]
and q are relativeprime.
The mapGq :
onRo, expressed by
satisfies the
required
twist condition at the boundaries. Indeed with 0 a = qw -[qw]
1 we conclude from(4.16):
Since
Gq
leaves aregular
measure invariant we conclude thatGq
possesses afixed
point
inRo.
One verifiesreadily
that itcorresponds
to aperiodic point
of
G,
hence oft,
with minimalperiod
q. This proves the first statement of Theorem 3.As for the second statement of Theorem 3 we observe that the invariant
curves found
by
J. Moser’s twist-theorem are in the closure of the set ofperiodic points.
This is well known and follows from anapplication
of the Birkoff-Lewis fixedpoint theorem,
see e.g.[15].
On the other hand it is also wellknown,
that the set E c A coveredby
the invariant curves contained in an annulus A fill out a set ofrelatively large
measurem(E) > (1 - E) - m(A)
with 0 E1,
we refer to J. Pbschel[16]
and the references therein.Applying
these observations to the annuliA : k p k + I
for alllarge integers
one sees that the set ofsubharmonic solutions of
(1.4)
is indeed of infiniteLebesgue
measure, sinceThis proves the statement of Theorem 3. ·
5. - Proof of Theorem 4
In case £ = 0 the
equation (1.7), expressed
in action andangle-variables (a, 8), (2.3) becomes,
after the transformation = Àa with a= -20132013:
n+2
and
fj being
continuous in t. One verifiesreadily
that the time 1 flow of thisequation
satisfieswith gj being analytic
and :SJ-L-ê
for p> p*(r, s).
Therefore theassumptions
of the twist-theorem are met and Theorem 4 followsreadily arguing
as at the very end of theproof
of Theorem 1.As for the Remark
(1.8)
we observe that inProposition
2 one loses onederivative in t at every successive iteration
step,
which is due to the derivative in theterm a s occurring
in the transformation formula. Thisexplains
our(9t .
g Psmoothness
requirement.
As for the smoothness in the x variable we should remark that the twist-theoremrequires merely C"16-small perturbations
for thoserotation numbers which we have considered above
[11],
and evenC3-small perturbations
for certain other irrational rotation numbers[12].
The work has been
supported by
theStiftung Volkswagenwerk.
REFERENCES
[1]
G.R. MORRIS, A caseof
boundedness in Littlewood’sproblem
onoscillatory differential
equations, Bull. Austral. Math. Soc., 14 (1976), pp. 71-93.[2]
J.E. LITTLEWOOD, Someproblems
in real andcomplex analysis,
(Heath,Lexington,
Mass. 1968).
[3]
J. MOSER, On invariant curvesof area-preserving
mappingsof
an annulus, Nachr.Akad. Wiss.
Göttingen
Math.Phys.
Kl. II, (1962), pp. 1-20.[4]
J. MOSER, Stable and random motions indynamical
systems, Ann. of Math. Studies 77, Princeton, N.J. (1973).[5]
P. HARTMAN, Onboundary
valueproblems for superlinear
second orderdifferential equations,
J. DifferentialEquations
26, (1977), pp. 37-53.[6]
H. JACOBOWITZ, Periodic solutionsof
x +f (t, x)
= 0 via thePoincaré-Birkoff fixed points
theorem, J. DifferentialEquations
20, (1976), pp. 37-53.[7]
K. SITNIKOV, Existenceof oscillating
motionsfor three-body problem,
Dokl. Akad.Nauk., SSSR., 133, 2, (1960), pp. 303-306.
[8]
V.M. ALEKSEEV,Quasirandom dynamical
systems, I, II, III, Math. USSR-Sb., 5, (1968), pp. 73-128; 6, (1968), pp. 505-560; 7, (1969), pp. 1-43.[9]
G.D. BIRKHOFF,Proof of
Poincaré’sgeometric
theorem, Trans. Amer. Math. Soc. 14, (1913), pp. 14-22.[10]
G.D. BIRKHOFF, An extensionof
Poincaré’s lastgeometric
theorem, Acta Math. 47, (1925).[11] H.
RÜSSMANN, On the existenceof
invariant curvesof
twistmappings of
an annulus,Preprint,
Mainz (1982).[12]
M. HERMAN, Démonstration du théorème des courbes translatées de nombre de rotation de type constant,manuscript,
Paris (1981).[13]
A. BAHRI - H. BERESTICKY, Forced Vibrationsof Superquadratic
HamiltonianSystems,
Acta. Math., 152, (1984), pp. 143-197.
[14]
M. HERMAN, Sur les courbes invariantes par lesdifféomorphismes
de l’anneau, Vol.1,
Astérisque
(1983), pp. 103-104, Vol. 2.Astérisque
(1986), pp. 144.[15]
E. ZEHNDER, Periodic solutionsof
Hamiltonianequations,
Lecture Notes in Math.,Springer,
1031, (1983), pp. 172-213.[16]
J. PÖSCHEL,Integrability of
HamiltonianSystems
on Cantor Sets, Comm. PureAppl.
Math. 36, (1982), pp. 653-695.
[17]
C.V. COFFMANN - D.F. ULLRICH, On the continuationof
solutionsof
a certainnon-linear
differential
equation, Monatsh. Math. 71, (1967), pp. 385-392.Ruhr-Universitat Bochum Mathematisches Institut Universitdtstrasse 150
D-4630 Bochum, West